MatrixDiagPartV3

Devuelve la parte diagonal por lotes de un tensor por lotes.

Devuelve un tensor con las diagonales `k[0]`-ésima a `k[1]`-ésima de la "entrada" por lotes.

Supongamos que la "entrada" tiene dimensiones "r" "[I, J, ..., L, M, N]". Sea `max_diag_len` la longitud máxima entre todas las diagonales que se extraerán, `max_diag_len = min(M + min(k[1], 0), N + min(-k[0], 0))` Sea `num_diags` Sea el número de diagonales a extraer, `num_diags = k[1] - k[0] + 1`.

Si `num_diags == 1`, el tensor de salida es de rango `r - 1` con forma `[I, J, ..., L, max_diag_len]` y valores:

diagonal[i, j, ..., l, n]
   = input[i, j, ..., l, n+y, n+x] ; if 0 <= n+y < M and 0 <= n+x < N,
     padding_value                 ; otherwise.
 

donde `y = max(-k [1], 0)`, `x = máx(k[1], 0)`.

De lo contrario, el tensor de salida tiene rango `r` con dimensiones `[I, J, ..., L, num_diags, max_diag_len]` con valores:

diagonal[i, j, ..., l, m, n]
   = input[i, j, ..., l, n+y, n+x] ; if 0 <= n+y < M and 0 <= n+x < N,
     padding_value                 ; otherwise.
 

donde `d = k[1] - m`, `y = max(-d, 0) - desplazamiento` y `x = max(d, 0) - desplazamiento`.

El "desplazamiento" es cero, excepto cuando la alineación de la diagonal es hacia la derecha.

offset = max_diag_len - diag_len(d) ; if (`align` in {RIGHT_LEFT, RIGHT_RIGHT}
                                            and `d >= 0`) or
                                          (`align` in {LEFT_RIGHT, RIGHT_RIGHT}
                                            and `d <= 0`)
          0                          ; otherwise
 

donde `diag_len(d) = min(cols - max(d, 0), filas + min(d, 0))`.

La entrada debe ser al menos una matriz.

Por ejemplo:

input = np.array([[[1, 2, 3, 4],  # Input shape: (2, 3, 4)
                    [5, 6, 7, 8],
                    [9, 8, 7, 6]],
                   [[5, 4, 3, 2],
                    [1, 2, 3, 4],
                    [5, 6, 7, 8]]])
 
 # A main diagonal from each batch.
 tf.matrix_diag_part(input) ==> [[1, 6, 7],  # Output shape: (2, 3)
                                 [5, 2, 7]]
 
 # A superdiagonal from each batch.
 tf.matrix_diag_part(input, k = 1)
   ==> [[2, 7, 6],  # Output shape: (2, 3)
        [4, 3, 8]]
 
 # A band from each batch.
 tf.matrix_diag_part(input, k = (-1, 2))
   ==> [[[0, 3, 8],  # Output shape: (2, 4, 3)
         [2, 7, 6],
         [1, 6, 7],
         [5, 8, 0]],
        [[0, 3, 4],
         [4, 3, 8],
         [5, 2, 7],
         [1, 6, 0]]]
 
 # LEFT_RIGHT alignment.
 tf.matrix_diag_part(input, k = (-1, 2), align="LEFT_RIGHT")
   ==> [[[3, 8, 0],  # Output shape: (2, 4, 3)
         [2, 7, 6],
         [1, 6, 7],
         [0, 5, 8]],
        [[3, 4, 0],
         [4, 3, 8],
         [5, 2, 7],
         [0, 1, 6]]]
 
 # max_diag_len can be shorter than the main diagonal.
 tf.matrix_diag_part(input, k = (-2, -1))
   ==> [[[5, 8],
         [9, 0]],
        [[1, 6],
         [5, 0]]]
 
 # padding_value = 9
 tf.matrix_diag_part(input, k = (1, 3), padding_value = 9)
   ==> [[[9, 9, 4],  # Output shape: (2, 3, 3)
         [9, 3, 8],
         [2, 7, 6]],
        [[9, 9, 2],
         [9, 3, 4],
         [4, 3, 8]]]